Abstract. Groundwater flow models are important tools in assessing baseline conditions and investigating management alternatives in groundwater systems. The usefulness of these models, however, is often hindered by insufficient knowledge regarding the magnitude and spatial distribution of the spatially-distributed parameters, such as hydraulic conductivity (K), that govern the response of these models. Proposed parameter estimation methods frequently are demonstrated using simplified aquifer representations, when in reality the groundwater regime in a given watershed is influenced by strongly-coupled surface-subsurface processes. Furthermore, parameter estimation methodologies that rely on a geostatistical structure of K often assume the parameter values of the geostatistical model as known or estimate these values from limited data.

In this study, we investigate the use of a data assimilation algorithm, the Ensemble Smoother, to provide enhanced estimates of K within a catchment system using the fully-coupled, surface-subsurface flow model CATHY. Both water table elevation and streamflow data are assimilated to condition the spatial distribution of K. An iterative procedure using the ES update routine, in which geostatistical parameter values defining the true spatial structure of K are identified, is also presented. In this procedure, parameter values are inferred from the updated ensemble of K fields and used in the subsequent iteration to generate the K ensemble, with the process proceeding until parameter values are converged upon. The parameter estimation scheme is demonstrated via a synthetic three-dimensional tilted v-shaped catchment system incorporating stream flow and variably-saturated subsurface flow, with spatio-temporal variability in forcing terms. Results indicate that the method is successful in providing improved estimates of the K field, and that the iterative scheme can be used to identify the geostatistical parameter values of the aquifer system. In general, water table data have a much greater ability than streamflow data to condition K. Future research includes applying the methodology to an actual regional study site.